No-code, vibe-code, same problems

A growing number of AI-pilled marketers have tried having an agent migrate their company's website away from a conventional CMS or site builder, looking to save money and take more control of their publishing setup. And while this can feel like an upgrade, in practice it's actually more of a lateral move, with these users hitting many of the same challenges and constraints as with Webflow, Framer, or other visual site builders.

Some things AI builders get right

To be clear, some gains people are seeing from AI-driven, self-serve web development are real. Using AI to build a site can save you money, especially if you already pay for an AI tool and can cancel a design tool or hosting subscription to help offset that cost.

Prompting a site into being can also be more pleasant and feel more direct than clicking and dragging sections around a canvas — even though agents plod through tasks more slowly, and with less precision, AI does let you achieve results by just saying what you want, though whether that works or not depends on what you're asking for and what model you're using.

Finally, performance scores can be a lot better with AI-created static sites than JavaScript-heavy Webflow or Framer sites. Some of this is due to AI builders' visual style being more minimalistic, and a truly cracked Webflow developer might be able to achieve similar results by stripping out visuals and optimizing everything. But a lot of improvements can come from agents generating more optimized code that loads and runs much faster. One marketer who wrote about their migration on LinkedIn reported raising their Lighthouse score from the mid-60s to 95 without a developer — impressive numbers in any event.

Beyond these, an AI-created site can feel like an upgrade by the mere fact that you (and AI) "built" it, and that your agent was able to use all its context and apparent intelligence to make a website that's better and more fit for purpose than what you could make with Webflow or Framer.

This is an 'upgrade' I'm more skeptical of. Just as social media can make people feel more connected to celebrities or brands when they aren't, really, AI chat interfaces can cause people to identify themselves with their output even when it turns out to look like most other agent-made projects.

This 'builder's high' from AI can trick you into taking on more responsibility for your site than you should, or realistically would want to. Day one makes you feel like Superman; by day ten, you might feel like your keyboard is made of Kryptonite.

Once the high wears off, you're left running a website. The hard parts of running one, like keeping it current, making it look like yours, and sharing the work with a team, don't go away because an agent wrote the code.

You're in control, for better or worse

Whether you're using AI or a visual canvas, once you take charge of building or migrating your site, all changes flow through you. This can be empowering, but also makes you the bottleneck for anything you want to change in your design or content.

AI agents can take on more of the heavy-feeling click labor: writing/editing copy, designing components and pages, building and publishing everything. But even if more of the low-level decision making is being handled by AI, just as with Webflow, the site only gets updated if you update it.

This is only a problem if or when the volume of changes you need to make on the regular starts to take enough of your time to feel onerous. A lot of websites just need gentle upkeep, especially early on, and being the sole content or technical owner may be such a light commitment that it feels fine, or even good, to have that much control.

AI or no AI, I think the standard here isn't to totally avoid these situations where you're the only site editor, but to leave yourself room to pivot whenever site management starts to become a chore rather than a pleasure.

One major pain point with large, long-running Webflow sites is that, being a visual editor with no find-and-replace, it can be challenging for new team members to understand their structure or make large-scale changes without a lot of clicking around and trying things. An agent can't necessarily build a future-proof content model or collaboration layer — in fact, you'll get better results if you don't burden the model with future requirements before you need them. But between having full visibility into your site's code and being able to leave behind context in the form of documentation, agents can leave you better positioned for an eventual migration or replatforming.

Making something distinctive is still hard

In fairness, if you're just starting a company and need to get a site out the door quickly, without either hiring or making website design your whole life, customizing a template is fine. Most people we work with start out there, and when you're hustling to get a new business off the ground it's hard to justify spending cycles on more than a bare minimum site.

A lot of Webflow sites (including ones our clients have relied on for years) start out from templates. What's more, component-based or section-based web design can often look generic and template-y even when professionally designed — it's not easy to make pages out of endlessly remixable blocks that don't obviously look like blocks.

AI-made sites have a similar problem — without a lot of prep work to create and document your design language, make assets like fonts and graphics available to agents, and so on, your design is likely to come out looking like a lot of other AI-designed sites.

This is expected — in the absence of prompts or skills to make a model deliver a specific result, LLMs tend to output an average of what's in the training data, resulting in generic, average-looking sites, full of elements that make sense individually but don't work when combined. If you want your brand to look like yours, you need your site to not look like an average of every other website — which, so far, requires a human touch whether you're using AI or Webflow.

Collaboration tends to break stuff

Webflow has notoriously only allowed one user at a time to edit pages in the Designer. The best-structured sites — built around CMS collections, components, and other good primitives — allow editors to easily make changes and even build new pages using components. Most Webflow sites aren't structured like that, and as sites and teams have grown, newcomers asked to take charge of a site can run into walls where they just can't publish what they need to publish.

With AI tools, whether collaboration is a problem and how those problems manifest varies by platform. Some products like Lovable or Replit have team plans and built-in collaborative tools, allowing multiple members of a team to prompt an agent to edit their site. If you built a site on your own laptop using Claude Code or Codex, you can take advantage of the same collaborative tools developers use (like GitHub pull requests), but that can force non-developers to deal with complexity they aren't accustomed to.

A system, not another tool

If replacing a no-code site with a vibe-coded one isn't an upgrade, what is?

Our clients who've moved past Webflow sites tend to hit a point where they need to make some kind of structural change — a new sitemap, adding new content collections, or a full-scale redesign or re-theming — that's hard to execute well within the constraints of a visual dev tool. With these kinds of big, strategic shifts, the issue is less about how the work is done than about deciding what specific changes will be most impactful and how to balance that work against ongoing updates to the current site.

These kinds of changes tend to accompany big shifts in who's responsible for a company's websites and content. A brand-new startup's founder or first marketing hire can take solo ownership of a Webflow site or AI-managed Astro codebase and stay on top of things. Later on, as more people join the marketing team and start to ramp up how often they put out new posts or pages, it becomes important to set up good systems that help keep everyone on the same page.

These aren't just software systems — AI can build those — but people systems, requiring careful thought about what kinds of content you're publishing, who's writing it, who's reviewing it, and how it makes its way from idea to publication.

When you hit these kinds of inflection points — or when you first start to feel like your day-one publishing stack is showing strain — companies like Bits&Letters can help you figure out where you're headed and how to get there with only the right amount of time, money, and rework.

If your site is starting to feel that strain, get in touch and we'll talk through where it should go next.